CN109164794B - Multivariable industrial process Fault Classification based on inclined F value SELM - Google Patents
Multivariable industrial process Fault Classification based on inclined F value SELM Download PDFInfo
- Publication number
- CN109164794B CN109164794B CN201811401207.4A CN201811401207A CN109164794B CN 109164794 B CN109164794 B CN 109164794B CN 201811401207 A CN201811401207 A CN 201811401207A CN 109164794 B CN109164794 B CN 109164794B
- Authority
- CN
- China
- Prior art keywords
- value
- data set
- formula
- variable
- fault
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
- G05B23/0205—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults
- G05B23/0218—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults
- G05B23/0243—Electric testing or monitoring by means of a monitoring system capable of detecting and responding to faults characterised by the fault detection method dealing with either existing or incipient faults model based detection method, e.g. first-principles knowledge model
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B2219/00—Program-control systems
- G05B2219/20—Pc systems
- G05B2219/24—Pc safety
- G05B2219/24065—Real time diagnostics
Abstract
The present invention relates to a kind of storehouse extreme learning machine Fault Classifications based on inclined F value SELM, are standardized first to training dataset；Secondly, finding out all fault types for the inclined F value of each variable, it is worth the weight of all variables according to the inclined F of all variables, further trained and test data is weighted；FSELM disaggregated model is constructed using storehouse extreme learning machine modeling method by training dataset；On this basis, test data set is standardized；The quality of the disaggregated model is verified finally by test data set.The present invention can be obviously improved performance of fault diagnosis, improve the accuracy rate of failure modes.
Description
Technical field
The invention belongs to industrial process fault diagnosis technology fields, are related to a kind of using based on the study of the inclined F value storehouse limit
The method that machine (referred to as: FSELM) classifies to multivariable industrial process failure.
Background technique
The progress of distributed computer control technology has greatly pushed the production-scale expansion of industrial process.For complexity
For huge industrial processes, efficient fault diagnosis technology not only facilitates the stability for improving device production, and
Also contribute to the personal safety of safeguard work personnel.The core of fault diagnosis technology is accurately to identify failure after failure occurs
Type, therefore failure modes technology is of great significance for field engineer's debugging.
Extreme learning machine (Extreme Learning Machine, referred to as: ELM) is by Guang-Bin Huang et al.
Successive ignition tune is not necessarily in a kind of single hidden layer feedforward neural network learning algorithm of proposition in 2006, the algorithm training process
It is whole, number, the suitable activation primitive of selection of hidden layer neuron need to be only reasonably selected, unique optimal solution can be obtained.Cause
It has many advantages, such as that training speed is fast, nicety of grading is high, generalization ability is strong, at present time series forecasting, image procossing,
The multiple fields such as failure modes are used widely.To handle amount industrial data complicated and changeable, researcher is further provided
Storehouse extreme learning machine (referred to as: SELM), constructs multilayer ELM under storehouse learning framework, to realize to data characteristic information
Depth is excavated and is extracted.However, traditional SELM puts on an equal footing all process variables in forming types classifier, assign
Identical weight, so that the significant variable information that part facilitates failure modes is covered by entire variable information, to reduce
Classifying quality.Therefore, how deterministic process variable difference of importance and assign different weights, become SELM fault grader structure
Challenge project in building.
Summary of the invention
The present invention, which does not excavate local significant variable information sufficiently for SELM fault grader, causes classifying quality to reduce
Problem provides a kind of multivariable industrial process Fault Classification based on inclined F value SELM.This method being capable of accurate extraction process
Facilitate the part significant variable of failure modes in data, reduces influence of the redundant variables to failure modes, improve failure modes
Accuracy.
In order to achieve the above object, the present invention provides a kind of multivariable industrial process failures based on inclined F value SELM point
Class method, contains following steps:
(1) industrial process normal operating floor data collection X is acquiredoWith C class fault condition data set { X1,X2,…XCMake
For training dataset, and use the mean value MX of normal operating floor data collectionoWith standard deviation SXoStandard is carried out to training dataset
Change processing, the normal operating floor data collection after being standardizedWith C class fault condition data set
(2) each fault condition data set, the inclined F value F (i, j) of calculating process variable, wherein F (i, j) expression the are directed to
I, 1≤i≤C class fault condition data setIn j-th of variable inclined F value；
(3) it sums to the inclined F value that all fault condition data sets obtain, obtains j-th of variable in all fault condition feelings
Overall F value Fs (j) partially under shape, and each variable is calculated according to accounting per (j) of the inclined F value of each variable in all variables
Importance weight w (j) in fault classification process；
(4) using importance weight w (j) to fault condition data setIt is weighted processing；
(5) FSELM disaggregated model is constructed using storehouse extreme learning machine modeling method, FSELM disaggregated model indicates are as follows: y
=f (x, W(1),…,W(P),V(1),…,V(P),B(1),…,B(P)), wherein x indicates input vector, i.e., process sample to be sorted
This, y indicates output vector, is that the classification of sample encodes, f () indicates a multilayer ELM storehouse network function, W(k), k=1,
2 ..., P indicates input layer weight parameter, V(k), k=1,2 ..., P indicate linear transformation battle array, B(k), k=1,2 ..., P,
Indicate output weight parameter, P indicates the storehouse number of plies；
(6) collecting test data set Xt, and place is standardized using the mean value and variance of normal operating floor data collection
Reason, the test data set after being standardized
(7) processing is weighted to test data set using importance weight w (j)；
(8) test data set after weighting is inputted into FSELM disaggregated model, is exported according to FSELM disaggregated model and determines event
Hinder type.
Further, the mean value MX of normal operating floor data collection is utilizedoWith standard deviation SXoBy formula (1) to normal behaviour
Make operating condition data set XoWith C class fault condition data set { X1,X2,…XCBe standardized, the expression formula of formula (1) are as follows:
Normal operating floor data collection XoWith C class fault condition data set { X1,X2,…XCStandardized by above-mentioned formula (1)
To get the normal operating floor data collection to after standardizing after processingWith C class fault condition data set
Further, in step (2), for fault condition data setJth variable is calculated by calculation formula (2)
Inclined F value F (i, j), calculation formula (2) indicate are as follows:
In formula, v=n0+ni- 2, n0For normal operative employee's condition data setIn number of samples, niFor fault condition data
CollectionIn number of samples, m be individual data collection process variable number, Ti 2To consider T when all process variables2Statistics
Amount indicates are as follows:
In formula, v=n0+ni- 2, n0For normal operative employee's condition data setIn number of samples, niFor fault condition data
CollectionIn number of samples, m be normal operative employee's condition data setWith fault condition data setThe variable of two datasets
Number, Ti 2To consider T when all process variables2Statistic indicates are as follows:
In formula,For b0Vector after removing j-th of variable,For biVector after removing j-th of variable；For Si
Matrix after removing jth row and jth column.
Further, it in step (3), sums to the inclined F value that all fault data collection obtain, obtains j-th of variable in institute
Overall F value Fs (j) partially under faulty situation, overall F value Fs (j) partially are indicated are as follows:
The overall accounting per (j) of the F value Fs (j) in all variables partially of j-th of variable, formula are calculated by formula (6)
(6) it indicates are as follows:
Each variable is calculated in fault classification process by formula (7) according to the inclined F value accounting per (j) of each variable
Importance weight w (j), formula (7) indicate are as follows:
In formula, m normal operating floor data collectionWith fault condition data setThe variable number of two datasets.
Further, in step (4), using importance weight w (j) by formula (8) to fault condition data setInto
Row weighting processing, formula (8) expression are as follows:
In formula,For weighting treated fault condition data set.
Further, in step (5), using storehouse extreme learning machine modeling method construct FSELM disaggregated model the step of
Are as follows:
(1) training dataset is constructedAnd corresponding output is worked out according to data category
Matrix Y；Hyper parameter P, L, L1 of storehouse extreme learning machine are set, and L, L >=1 indicate implicit layer unit number, L1,1≤L1≤C table
Column dimension after showing output weight matrix dimensionality reduction；
(2) random initializtion input layer weight parameter W(k), k=1,2 ..., P, wherein W(1)Matrix, W are tieed up for L × m(2)... W(P)Matrix is tieed up for (L-L1) × m；Initialize hidden layer serial number p=1；
(3) p-th of hidden layer output matrix H is calculated(p)=g (X (W(p))T), wherein g () indicates that sigmoid activates letter
Number；
(4) output weight matrix B is calculated(p)=(H(p))*Y, wherein B(p)Matrix is tieed up for L × C,*Indicate MP generalized inverse fortune
It calculates；
(5) to output weight matrix B(p)Pivot analysis is carried out, the linear transformation battle array V of L × L1 dimension is obtained(p)；
(6)+1 hidden layer matrix H of pth is found out(p+1)=[H(p)V(p) g(X(W(p+1))T)]；
(7) p=p+1 is enabled, if p < P, by new hidden layer matrix H(p+1)Instead of H(p), step (4)-(6) are repeated, into one
Step calculates B(p+1)、V(p+1)；Otherwise reach the storehouse number of plies, FSELM model construction is completed；The FSELM disaggregated model of building indicates
Are as follows:
Y=f (x, W(1),…,W(P),V(1),…,V(P),B(1),…,B(P)) (9)
Further, in step (6), the mean value MX of normal operating floor data collection is utilizedoWith standard deviation SXoPass through formula
(10) to test data set XtIt is standardized, formula (10) indicates are as follows:
Test data set XtBy after above-mentioned formula (10) standardization to get to standardization after test data set
Further, in step (7), using importance weight w (j) by formula (11) to test data setIt carries out
Weighting processing, formula (11) indicate are as follows:
In formula,For weighting treated test data set.
Further, in step (8), by the test data set after weightingFSELM disaggregated model is inputted, according to FSELM
Disaggregated model output obtains test data setFault type matrixFault type matrixIt indicates are as follows:
Compared with prior art, the beneficial effects of the present invention are:
Multivariable industrial process Fault Classification provided by the invention based on inclined F value SELM, to the training number of acquisition
It is standardized according to collection, finds out all fault types for the inclined F value of each variable, according to the inclined F value of all variables
It obtains the weight of all variables, further trained and test data is weighted, storehouse pole is utilized by training dataset
It limits learning machine modeling method and constructs FSELM disaggregated model, verify the quality of the disaggregated model by test data set, and according to
FSELM disaggregated model classifies to fault type.The present invention excavates the process variable for facilitating failure modes and gives larger
Weight facilitates the part significant variable of failure modes in accurate extraction process data, reduces redundant variables to failure modes
It influences, the accuracy rate of failure modes can be effectively improved.
Detailed description of the invention
Fig. 1 is the flow chart of the multivariable industrial process Fault Classification of the present invention based on inclined F value SELM；
Fig. 2 is the structure chart of Tennessee-Yi Siman described in the embodiment of the present invention (referred to as: TE) process；
Fig. 3 a is the inclined F value of each variable of failure 1 described in the embodiment of the present invention；
Fig. 3 b is the inclined F value of each variable of failure 2 described in the embodiment of the present invention；
Fig. 3 c is the inclined F value of each variable of failure 3 described in the embodiment of the present invention；
Fig. 3 d is the inclined F value of each variable of failure 4 described in the embodiment of the present invention.
Specific embodiment
In the following, the present invention is specifically described by illustrative embodiment.It should be appreciated, however, that not into one
In the case where step narration, element, structure and features in an embodiment can also be advantageously incorporated into other embodiments
In.
Referring to Fig. 1, present invention discloses a kind of multivariable industrial process Fault Classifications based on inclined F value SELM, contain
There are following steps:
(1) industrial process normal operating floor data collection X is acquiredoWith C class fault condition data set { X1,X2,…XCMake
For training dataset, and use the mean value MX of normal operating floor data collectionoWith standard deviation SXoBy formula (1) to training data
Collection is standardized, the expression formula of formula (1) are as follows:
Training dataset by after above-mentioned formula (1) standardization to get to standardization after normal operating floor data
CollectionWith C class fault condition data set
(2) it is directed to fault condition data setThe inclined F value F (i, j) of jth variable is calculated by calculation formula (2),
Middle F (i, j) indicates the i-th class fault condition data setIn j-th of variable inclined F value.Calculation formula (2) indicates are as follows:
In formula, v=n0+ni- 2, n0For normal operative employee's condition data setIn number of samples, niFor fault condition data
CollectionIn number of samples, m be individual data collection process variable number, Ti 2To consider T when all process variables2Statistics
Amount indicates are as follows:
In formula, v=n0+ni- 2, n0For normal operative employee's condition data setIn number of samples, niFor fault condition data
CollectionIn number of samples, m be normal operative employee's condition data setWith fault condition data setThe variable of two datasets
Number, Ti 2To consider T when all process variables2Statistic indicates are as follows:
In formula,For b0Vector after removing j-th of variable,For biVector after removing j-th of variable；For Si
Matrix after removing jth row and jth column.
If the inclined F value for certain class fault data is bigger, illustrate that contribution when this variable classifies to such failure is got over
Greatly；Conversely, the inclined F value for certain class fault data is smaller, the contribution for illustrating that this variable classifies to such failure is smaller.
(3) it sums to the inclined F value that all fault data collection obtain, it is total under all failure situations to obtain j-th of variable
The inclined F value Fs (j) of body, overall F value Fs (j) partially are indicated are as follows:
The overall accounting per (j) of the F value Fs (j) in all variables partially of j-th of variable, formula are calculated by formula (6)
(6) it indicates are as follows:
Each variable is calculated in fault classification process by formula (7) according to the inclined F value accounting per (j) of each variable
Importance weight w (j), formula (7) indicate are as follows:
(4) pass through formula (8) to fault condition data set using importance weight w (j)It is weighted processing, formula
(8) it expresses are as follows:
In formula,For weighting treated fault condition data set.
(5) FSELM disaggregated model is constructed using storehouse extreme learning machine modeling method, FSELM disaggregated model indicates are as follows: y
=f (x, W(1),…,W(P),V(1),…,V(P),B(1),…,B(P)), wherein x indicates input vector, i.e., process sample to be sorted
This, y indicates output vector, is that the classification of sample encodes, f () indicates a multilayer ELM storehouse network function, W(k), k=1,
2 ..., P indicates input layer weight parameter, V(k), k=1,2 ..., P indicate linear transformation battle array, B(k), k=1,2 ..., P,
Indicate output weight parameter.Construct the specific steps of FSELM disaggregated model are as follows:
(1) training dataset is constructedAnd corresponding output is worked out according to data category
Matrix Y；Hyper parameter P, L, L1 of storehouse extreme learning machine are set, and L, L >=1 indicate implicit layer unit number, L1,1≤L1≤C table
Column dimension after showing output weight matrix dimensionality reduction；
(2) random initializtion input layer weight parameter W(k), k=1,2 ..., P, wherein W(1)Matrix, W are tieed up for L × m(2)... W(P)Matrix is tieed up for (L-L1) × m；Initialize hidden layer serial number p=1；
(3) p-th of hidden layer output matrix H is calculated(p)=g (X (W(p))T), wherein g () indicates that sigmoid activates letter
Number；
(4) output weight matrix B is calculated(p)=(H(p))*Y, wherein B(p)Matrix is tieed up for L × C,*Indicate MP generalized inverse fortune
It calculates；
(5) to output weight matrix B(p)Pivot analysis is carried out, the linear transformation battle array V of L × L1 dimension is obtained(p)；
(6)+1 hidden layer matrix H of pth is found out(p+1)=[H(p)V(p) g(X(W(p+1))T)]；
(7) p=p+1 is enabled, if p < P, by new hidden layer matrix H(p+1)Instead of H(p), step (4)-(6) are repeated, into one
Step calculates B(p+1)、V(p+1)；Otherwise reach the storehouse number of plies, FSELM model construction is completed；The FSELM disaggregated model of building indicates
Are as follows:
Y=f (x, W(1),…,W(P),V(1),…,V(P),B(1),…,B(P)) (9)
(6) collecting test data set Xt, utilize the mean value MX of normal operating floor data collectionoWith standard deviation SXoPass through public affairs
Formula (10) is to test data set XtIt is standardized, formula (10) indicates are as follows:
Test data set XtBy after above-mentioned formula (10) standardization to get to standardization after test data set
(7) pass through formula (11) to test data set using importance weight w (j)It is weighted processing, formula (11)
It indicates are as follows:
In formula,For weighting treated test data set.
(8) test data set after weighting is inputted into FSELM disaggregated model, survey is obtained according to the output of FSELM disaggregated model
Try data setFault type matrixFault type matrixIt indicates are as follows:
In the above method, step (1) to (five) is the off-line modeling stage, and step (6) to (eight) is the on-line testing stage.
The above-mentioned Fault Classification of the present invention is standardized training dataset；Find out all fault types for
The inclined F value of each variable is worth the weight of all variables according to the inclined F of all variables, further to training and test
Data are weighted；On this basis, test data set is standardized；It is established by training data and is based on storehouse pole
Limit the failure modes model of learning machine.The present invention excavates the process variable for facilitating failure modes and gives greater weight, accurately
Facilitate the part significant variable of failure modes in extraction process data, reduces influence of the redundant variables to failure modes, it can
Effectively improve the accuracy rate of failure modes.
In order to be illustrated more clearly that the beneficial effect of the above-mentioned Fault Classification of the present invention, below in conjunction with specific embodiment
Further explanation is made to the above-mentioned Fault Classification of the present invention.
Embodiment: Tennessee-Yi Siman (hereinafter referred to as: TE) process be by Yisiman Chemical Company, the U.S. Downs and
The experiment porch that Vogel is established according to an actual chemical process, is now widely used in access control algorithm and process monitoring
The superiority and inferiority of method.Referring to fig. 2, TE process is mainly made of five units, including reactor, product condenser, gas-liquid separator,
Recycle compressor and stripper composition.TE process totally 53 variables, including 22 continuous measurands, 19 component variables
With 12 performance variables.Wherein 34 variables include 22 continuous measurands and 12 performance variables, as shown in table 1.
Table 1
| Variable label | Variable description | Variable label | Variable description |
| 1 | A charging (stream 1) | 18 | Stripper temperature |
| 2 | D charging (stream 2) | 19 | Stripper flow |
| 3 | E charging (stream 3) | 20 | Compressor horsepower |
| 4 | A and C charging (stream 4) | 21 | Reactor cooling water outlet temperature |
| 5 | Recirculating mass (stream 8) | 22 | Separator cooling water outlet temperature |
| 6 | Reactor feed speed (stream 6) | 23 | D inlet amount (stream 2) |
| 7 | Reactor pressure | 24 | E inlet amount (stream 3) |
| 8 | Reactor grade | 25 | A inlet amount (stream 1) |
| 9 | Temperature of reactor | 26 | A and C doses (stream 4) |
| 10 | It is vented rate | 27 | Compressor recycle valve |
| 11 | Product separator temperature | 28 | Blow valve |
| 12 | Product separator liquid level | 29 | Separator pot flow quantity (stream 10) |
| 13 | Product separator pressure | 30 | Stripper liquid product flow (stream 11) |
| 14 | Product separator tower bottom flow (stream 10) | 31 | Stripper water flow valve |
| 15 | Stripper grade | 32 | Reactor cold water flow |
| 16 | Stripper pressure | 33 | Condenser cold water flow |
| 17 | Stripper column underflow amount (stream 11) | 34 | Mixing speed (nothing) |
4 class fault datas of TE procedure fault are used in experiment, it is as shown in table 2, accurate for the classification of verifying each method
Rate.
Table 2
In this embodiment, mainly the failure 1 of TE process, failure 2, failure 3, failure 4 this 4 class fault type are divided
Class, wherein every each 300 samples of class failure in training set, every each 200 samples of class failure in test set.
What Fig. 3 a was indicated is the inclined F value of each variable in failure 1, and that Fig. 3 b is indicated is the inclined F of each variable in failure 2
Value, what Fig. 3 c was indicated is the inclined F value of each variable in failure 3, and what Fig. 3 d was indicated is the inclined F value of each variable in failure 4.According to
The accounting of the inclined F value of each variable and the inclined F value of each variable in all variables show that the weight of 15 variables therein is
1, i.e. variable 1,44,50,40,9,17,52,21,38,18,35,41,34,29,10.
It is utilized respectively that extreme learning machine (referred to as: ELM), storehouse extreme learning machine (referred to as: SELM), the present invention is based on inclined F
The multivariable industrial process Fault Classification (referred to as: FSELM) of value SELM classifies to the above 4 class TE failure.Failure point
The results are shown in Table 3 for class accuracy rate.
Table 3
| Failure 1 | Failure 2 | Failure 3 | Failure 4 | Average value | |
| ELM | 0.6662 | 0.5858 | 0.5458 | 0.9160 | 0.6785 |
| SELM | 0.8598 | 0.6110 | 0.5558 | 0.9432 | 0.7425 |
| FSELM | 0.9925 | 0.7180 | 0.6535 | 0.9655 | 0.8324 |
By above-mentioned table 3 it is found that proposed by the present invention can significantly improve classification based on inclined F value SELM Fault Classification
Accuracy rate.
Embodiment provided above only with illustrating the present invention for convenience, and it is not intended to limit the protection scope of the present invention, In
Technical solution scope of the present invention, person of ordinary skill in the field make various simple deformations and modification, should all include
In the above claim.
Claims (9)
1. a kind of multivariable industrial process Fault Classification based on inclined F value SELM, contains following steps:
(1) industrial process normal operating floor data collection X is acquiredoWith C class fault condition data set { X1,X2,…XCAs training
Data set, and use the mean value MX of normal operating floor data collectionoWith standard deviation SXoTraining dataset is standardized,
Normal operating floor data collection after being standardizedWith C class fault condition data set
(2) it is directed to each fault condition data set, the inclined F value F (i, j) of calculating process variable, wherein F (i, j) indicates i-th, 1
≤ i≤C class fault condition data setIn j-th of variable inclined F value；
(3) it sums to the inclined F value that all fault condition data sets obtain, obtains j-th of variable under all fault condition situations
Overall F value Fs (j) partially, and each variable is calculated former according to accounting per (j) of the inclined F value of each variable in all variables
Hinder the importance weight w (j) in assorting process；
(4) using importance weight w (j) to fault condition data setIt is weighted processing；
(5) FSELM disaggregated model is constructed using storehouse extreme learning machine modeling method, FSELM disaggregated model indicates are as follows: y=f
(x,W(1),…,W(P),V(1),…,V(P),B(1),…,B(P)), wherein x indicates input vector, i.e., process sample to be sorted, y
It indicates output vector, is that the classification of sample encodes, f () indicates a multilayer ELM storehouse network function, W(k), k=1,
2 ..., P indicates input layer weight parameter, V(k), k=1,2 ..., P indicate linear transformation battle array, B(k), k=1,2 ..., P,
Indicate output weight parameter, P indicates the storehouse number of plies；
(6) collecting test data set Xt, and be standardized using the mean value and variance of normal operating floor data collection, it obtains
Test data set after to standardization
(7) processing is weighted to test data set using importance weight w (j)；
(8) test data set after weighting is inputted into FSELM disaggregated model, is exported according to FSELM disaggregated model and determines failure classes
Type.
2. the multivariable industrial process Fault Classification as described in claim 1 based on inclined F value SELM, which is characterized in that
In step (1), the mean value MX of normal operating floor data collection is utilizedoWith standard deviation SXoBy formula (1) to training dataset into
Row standardization, the expression formula of formula (1) are as follows:
Training dataset by after above-mentioned formula (1) standardization to get to standardization after normal operating floor data collection
With C class fault condition data set
3. the multivariable industrial process Fault Classification as claimed in claim 2 based on inclined F value SELM, which is characterized in that
In step (2), for fault condition data setThe inclined F value F (i, j) of jth variable is calculated by calculation formula (2), is calculated
Formula (2) indicates are as follows:
In formula, v=n0+ni- 2, n0For normal operative employee's condition data setIn number of samples, niFor fault condition data set
In number of samples, m be individual data collection process variable number, Ti 2To consider T when all process variables2Statistic, table
It is shown as:
In formula, b0For the mean vector of normal class data；biIt is the mean vector of the i-th class fault data；SiFor combinatorial matrixCovariance matrix；To reject the T after variable ji 2Value, it may be assumed that
In formula,For b0Vector after removing j-th of variable,For biVector after removing j-th of variable；For SiRemove
Matrix after jth row and jth column.
4. the multivariable industrial process Fault Classification as claimed in claim 3 based on inclined F value SELM, which is characterized in that
In step (3), sums to the inclined F value that all fault data collection obtain, it is total under all failure situations to obtain j-th of variable
The inclined F value Fs (j) of body, overall F value Fs (j) partially are indicated are as follows:
The overall accounting per (j) of the F value Fs (j) in all variables partially of j-th of variable, formula (6) are calculated by formula (6)
It indicates are as follows:
It is important in fault classification process by each variable of formula (7) calculating according to the inclined F value accounting per (j) of each variable
Property weight w (j), formula (7) indicate are as follows:
5. the multivariable industrial process Fault Classification as claimed in claim 4 based on inclined F value SELM, which is characterized in that
In step (4), using importance weight w (j) by formula (8) to fault condition data setIt is weighted processing, formula
(8) it expresses are as follows:
In formula,For weighting treated fault condition data set.
6. the multivariable industrial process Fault Classification as claimed in claim 5 based on inclined F value SELM, which is characterized in that
In step (5), using storehouse extreme learning machine modeling method construct FSELM disaggregated model the step of are as follows:
(1) training dataset is constructedAnd corresponding output matrix Y is worked out according to data category；
Hyper parameter P, L, L1 of storehouse extreme learning machine are set, and L, L >=1 indicate implicit layer unit number, L1,1≤L1≤C expression output
Column dimension after weight matrix dimensionality reduction；
(2) random initializtion input layer weight parameter W(k), k=1,2 ..., P, wherein W(1)Matrix, W are tieed up for L × m(2)... W(P)Matrix is tieed up for (L-L1) × m；Initialize hidden layer serial number p=1；
(3) p-th of hidden layer output matrix H is calculated(p)=g (X (W(p))T), wherein g () indicates sigmoid activation primitive；
(4) output weight matrix B is calculated(p)=(H(p))*Y, wherein B(p)Matrix is tieed up for L × C,*Indicate the inverse operation of MP broad sense；
(5) to output weight matrix B(p)Pivot analysis is carried out, the linear transformation battle array V of L × L1 dimension is obtained(p)；
(6)+1 hidden layer matrix H of pth is found out(p+1)=[H(p)V(p)g(X(W(p+1))T)]；
(7) p=p+1 is enabled, if p < P, by new hidden layer matrix H(p+1)Instead of H(p), repeat step (4)-(6), further to count
Calculate B(p+1)、V(p+1)；Otherwise reach the storehouse number of plies, FSELM model construction is completed；The FSELM disaggregated model of building indicates are as follows:
Y=f (x, W(1),…,W(P),V(1),…,V(P),B(1),…,B(P)) (9)。
7. the multivariable industrial process Fault Classification as claimed in claim 6 based on inclined F value SELM, which is characterized in that
In step (6), the mean value MX of normal operating floor data collection is utilizedoWith standard deviation SXoBy formula (10) to test data set
XtIt is standardized, formula (10) indicates are as follows:
Test data set XtBy after above-mentioned formula (10) standardization to get to standardization after test data set
8. the multivariable industrial process Fault Classification as claimed in claim 7 based on inclined F value SELM, which is characterized in that
In step (7), using importance weight w (j) by formula (11) to test data setIt is weighted processing, formula (11)
It indicates are as follows:
In formula,For weighting treated test data set.
9. the multivariable industrial process Fault Classification as claimed in claim 8 based on inclined F value SELM, which is characterized in that
In step (8), by the test data set after weightingFSELM disaggregated model is inputted, is obtained according to the output of FSELM disaggregated model
Test data setFault type matrixFault type matrixIt indicates are as follows:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811401207.4A CN109164794B (en) | 2018-11-22 | 2018-11-22 | Multivariable industrial process Fault Classification based on inclined F value SELM |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811401207.4A CN109164794B (en) | 2018-11-22 | 2018-11-22 | Multivariable industrial process Fault Classification based on inclined F value SELM |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109164794A CN109164794A (en) | 2019-01-08 |
| CN109164794B true CN109164794B (en) | 2019-11-29 |
Family
ID=64875111
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811401207.4A Active CN109164794B (en) | 2018-11-22 | 2018-11-22 | Multivariable industrial process Fault Classification based on inclined F value SELM |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109164794B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110490496B (en) * | 2019-06-19 | 2022-03-11 | 中南大学 | Method for screening sensitive variables influencing product quality in complex industrial process based on stepwise reduction |
| CN110244692B (en) * | 2019-06-20 | 2020-07-07 | 中国石油大学(华东) | Chemical process micro-fault detection method |
| CN113159225B (en) * | 2021-05-17 | 2022-04-19 | 中国石油大学(华东) | Multivariable industrial process fault classification method |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105182955B (en) * | 2015-05-15 | 2016-06-22 | 中国石油大学(华东) | A kind of multivariate industrial process fault recognition method |
| CN108052793B (en) * | 2017-12-26 | 2023-10-31 | 杭州电子科技大学 | Mobile pollution source emission concentration prediction method based on fuzzy weighted ELM |
-
2018
- 2018-11-22 CN CN201811401207.4A patent/CN109164794B/en active Active
Non-Patent Citations (1)
| Title |
|---|
| ASELM: Adaptive semi-supervised ELM with application in question subjectivity identification;Hongping Fu et al.;《Neurocomputing》;20160522;第599-609页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109164794A (en) | 2019-01-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106355030B (en) | A kind of fault detection method based on analytic hierarchy process (AHP) and Nearest Neighbor with Weighted Voting Decision fusion | |
| CN106371427B (en) | Industrial process Fault Classification based on analytic hierarchy process (AHP) and fuzzy Fusion | |
| Fu et al. | Fault diagnosis for rolling bearings based on composite multiscale fine-sorted dispersion entropy and SVM with hybrid mutation SCA-HHO algorithm optimization | |
| CN106845717B (en) | Energy efficiency evaluation method based on multi-model fusion strategy | |
| CN111626336B (en) | Subway fault data classification method based on unbalanced data set | |
| CN106843195B (en) | The Fault Classification differentiated based on adaptive set at semi-supervised Fei Sheer | |
| CN109164794B (en) | Multivariable industrial process Fault Classification based on inclined F value SELM | |
| CN104712542B (en) | A kind of reciprocating compressor sensitive features based on Internet of Things are extracted and method for diagnosing faults | |
| Arbin et al. | Comparative analysis between k-means and k-medoids for statistical clustering | |
| CN106202952A (en) | A kind of Parkinson disease diagnostic method based on machine learning | |
| CN106649789A (en) | Integrated semi-supervised Fisher's discrimination-based industrial process fault classifying method | |
| CN109740655B (en) | Article scoring prediction method based on matrix decomposition and neural collaborative filtering | |
| CN116108758B (en) | Landslide susceptibility evaluation method | |
| CN110381079A (en) | Network log method for detecting abnormality is carried out in conjunction with GRU and SVDD | |
| CN106681305A (en) | Online fault diagnosing method for Fast RVM (relevance vector machine) sewage treatment | |
| CN108921230A (en) | Method for diagnosing faults based on class mean value core pivot element analysis and BP neural network | |
| CN108875772A (en) | A kind of failure modes model and method being limited Boltzmann machine and intensified learning based on the sparse Gauss Bernoulli Jacob of stacking | |
| CN107704883A (en) | A kind of sorting technique and system of the grade of magnesite ore | |
| CN112766283B (en) | Two-phase flow pattern identification method based on multi-scale convolution network | |
| CN111126820A (en) | Electricity stealing prevention method and system | |
| CN110021340A (en) | A kind of RNA secondary structure generator and its prediction technique based on convolutional neural networks and planning dynamic algorithm | |
| CN108491991A (en) | Constraints analysis system based on the industrial big data product duration and method | |
| CN109816030A (en) | A kind of image classification method and device based on limited Boltzmann machine | |
| CN113159225B (en) | Multivariable industrial process fault classification method | |
| CN110378035A (en) | It is a kind of that soft-measuring modeling method is hydrocracked based on deep learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |